Accelerated gradient learning algorithm for neural network weights update

This work proposes decomposition of square approximation algorithm for neural network weights update. Suggested improvement results in alternative method that converge in less iteration and is inherently parallel. Decomposition enables parallel execution convenient for implementation on computer grid. Improvements are reflected in accelerated learning rate which may be essential for time critical decision processes. Proposed solution is tested and verified on multilayer perceptrons neural network case study, varying a wide range of parameters, such as number of inputs/outputs, length of input/output data, number of neurons and layers. Experimental results show time savings up to 40% in multiple thread execution.     

Error in measurement due to stylus kinematics

A study of the error in measurement due to the kinematics of the stylus of the measuring instrument is presented. A method of determining the point of separation of the stylus from the measured surface and the path of the stylus after separation is established. Stylus spring stiffness to prevent separation of the stylus and to attain a minimum measuring force is also established.     

Measurement of the optical parameters of the Virgo interferometer

The Virgo interferometer, aimed at detecting gravitational waves, is now in a commissioning phase. Measurements of its optical properties are needed for the understanding of the instrument. We present the techniques developed for the measurement of the optical parameters of Virgo. These parameters are compared with the Virgo specifications.     

Handling mixed sets of tasks in combined offline and online scheduled real-time systems

Many industrial applications with real-time demands are composed of mixed sets of tasks with a variety of requirements. These can be in the form of standard timing constraints, such as period and deadline, or complex, e.g., to express application specific or nontemporal constraints, reliability, performance, etc. As many algorithms focus on specific sets of task types and constraints only, system design has to focus on those supported by a particular algorithm, at the expense of the rest. In this paper, we present a method to deal with a combination of mixed sets of tasks and constraints: periodic tasks with complex and simple constraints, soft and firm aperiodic, and sporadic tasks. We propose the use of an offline scheduler to manage complex timing and resource constraints of periodic tasks and transform these into a simple EDF model with start-times and deadlines. At run-time, the execution of the offline scheduled tasks is flexibly shifted in order to allow for feasible inclusion of dynamically arriving sporadic and aperiodic tasks. Sporadic tasks are guaranteed offline based on their worst-case activation frequencies. At run-time, this pessimism is reduced by the online algorithm which uses the exact knowledge about sporadic arrivals to reclaim resources and improve response times and acceptance of firm aperiodic tasks.     

Green Nanocoatings Prepared by Crosslinking Self-Assembled Fatty Acids on Metals

Self-assembled monolayers (SAMs) are an important element of modern nanotechnology and surface functionalization. However, their application is still limited because they are easily removed from the surface of the object in corrosive environments. Crosslinking would make SAMs more resistant to the corrosive environment they are exposed to. In this work, how to strongly crosslink SAMs made of non-toxic and biodegradable fatty acids on metal surfaces using ionizing radiation has been demonstrated for the first time. The crosslinked nanocoatings are stable over time and have significantly improved properties compared to SAMs. Thus, crosslinking opens up the possibility of using SAMS in a variety of different systems and on different materials for surface functionalization to achieve stable and durable surface properties such as biocompatibility or selective reactivity.